The Agent Skills Directory

[SAFE]: All network operations are conducted with the vendor's official domains, api.novoads.ai and novoads.ai, which is required for the skill's primary media generation services.
[COMMAND_EXECUTION]: The skill uses system tools including curl, ffmpeg, ffprobe, and Whisper to facilitate API communication, analyze media file properties, and perform automated transcription for video QA. These actions are transparently documented and aligned with the skill's functional requirements.
[EXTERNAL_DOWNLOADS]: The skill downloads generated MP4 and image files from the vendor's API to a local outputs directory. These downloads originate from the verified vendor domain and are intended for user review.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted user input such as product descriptions and ad scripts to construct generation prompts. While boundary markers are not explicitly enforced, the resulting data is sent to a specific vendor API and analyzed by local tools, minimizing the risk of harmful cross-skill side effects.
Ingestion points: SKILL.md and prompting/guide.md (via user input for ad concepts).
Boundary markers: Absent; user text is interpolated into prompt templates.
Capability inventory: curl (network), ffmpeg/ffprobe (media analysis), whisper (transcription).
Sanitization: Not explicitly detailed, but media output verification serves as a logical check.
[SAFE]: Credentials are managed securely. The skill instructs users to store API keys in a gitignored .env file and explicitly warns the agent against printing keys or logging sensitive Authorization headers.

novoads-api